Anthropic launches a public beta on Messages API of Message Threads，Claude Code: time to first Token for long conversations falls by approximately 35%
BIBIBI
AT A GLANCE
The news flash reports that Anthropic is publicly testing message threads,Claude Code reducing data transmission and time to first Token for long conversations.
Article
Claude Code Long conversations: time to first Token shortened 35%, request payload reduced 99.8%According to a Beating AI news flash, Anthropic has launched a public beta of Messages API on Message Threads(message threads), reducing the overhead of AI Agent repeatedly uploading conversation history.
Previously, each time a Coding Agent called Claude, it had to resend earlier messages and tool execution results. Now, it only needs to upload the full conversation once; subsequent turns send the new message and the previous response's ID. The server temporarily stores the history, saving time on repeated transmission, validation, and tokenization.
The feature is enabled by default in Claude Code. Anthropic disclosed that, for requests ranging from 500000 to 1000000 Token, median time to first Token fell by approximately 35%. For long conversations, the median request payload shrank from approximately 800KB to 1.8KB, a reduction of approximately 99.8%.
However, the model must still process the full context, so Token usage and costs remain unchanged. Message threads stored on the server also expire after 5 minutes of inactivity, at which point the history must be uploaded again.
A Beating AI news flash reports that Anthropic has launched a public beta on Messages API of Message Threads.
02
Previously, each time a Coding Agent called Claude, it had to resend earlier messages and tool execution results.
03
With the feature enabled, the full conversation is uploaded initially; subsequent turns send only the new message and the previous response's ID。
04
The server temporarily stores the history to reduce the time required for repeated transmission, validation, and tokenization.
05
The feature is enabled by default in Claude Code.
06
Anthropic disclosed that, for requests ranging from 500000 to 1000000 Token, median time to first Token fell by approximately 35%。
07
For long conversations, the median request payload shrank from approximately 800KB to 1.8KB, a reduction of approximately 99.8%。
08
The model must still process the full context, so Token usage and costs remain unchanged.
09
Message threads stored on the server expire after 5 minutes of inactivity, after which the history must be uploaded again.
AI-assisted interpretation
The following is analysis, separate from reported facts. Verify important claims independently.
This feature temporarily stores long conversation histories on the server. After the full content is sent initially, subsequent requests continue the thread using the response ID, without repeatedly uploading the same history. It primarily reduces network transmission and the wait before the first Token appears, but does not reduce the context the model actually processes.
Why it matters to readers
For long-running coding agents, smaller request payloads and shorter wait times could make interactions faster and reduce the overhead of repeatedly uploading large conversation histories.
Beginners can think of it this way: the system remembers the preceding conversation, so only new content needs to be sent afterward; however, the Token usage used as the basis for billing does not decrease.
Risks and unknowns
The evidence comes from a media news flash and does not include a firsthand announcement from Anthropic. The scope of default enablement and the performance figures still require cross-checking against official materials.
The figures in the article apply to requests ranging from 500000 to 1000000 Token. Whether requests of other lengths or other workflows can achieve the same improvements is unknown.
Message threads expire after 5 minutes of inactivity. The frequency and impact of re-uploading history in actual use are unknown.
Related Developments
Loading event timeline…
Related concepts
Token
This term is not in the glossary yet. Browse related concepts in the glossary.
